AI and me
I'm a data scientist in aerospace, passionate about AI and new technology. I've been using agentic AI heavily since December 2025. In a few months I didn't just add one more assistant: I built my own ecosystem of tools around it, so it works the way I work. I'm starting this blog to share my experiments and ideas.
I quickly stopped seeing AI as an answer engine. I treat it as a colleague I delegate whole tasks to — but a colleague under contract, with clear rules. First, autonomy: when a task is clear, it gets done, without asking permission at every step; I'm interrupted only for a real decision. Then, opinion: when I ask a question, I expect a reasoned recommendation, not five equivalent options. Finally, rigor: nothing ships until the hypothesis has survived an attempt to refute it. Facing a bug, I don't guess the cause, I measure it. Measure, falsify, reformulate. Ideas live or die by data, not by argument.
To make all this hold, I built a frame. The AI doesn't wander: it follows task pipelines with checkpoints that refuse to let mediocre work through, it inspects the real structure of the code instead of guessing it, and every test must actually test something. The principle is simple: the clearer the frame, the safer the autonomy. Freedom without structure breeds chaos; structure without freedom breeds slowness. I want both at once.
What I take away is that AI didn't replace my judgment — it amplified it. Poorly framed, it amplifies confusion; well framed, it amplifies rigor. The future belongs neither to those who reject AI, nor to those who delegate to it blindly, but to those who can make it a demanding collaborator. That is, in the end, exactly what I expect from a good human colleague.
Latest posts
- No telescope will save Trisolaris (or your agent) What the three-body problem teaches about the prediction limits of agentic AI, and why the only strategy that depends neither on model generation nor budget is to constrain the dynamics.
- The RNA moment of agentic AI What the origin of life teaches us about self-improvement loops — and why it's the constraint, not the supervisor, that closes the loop.
- What you won't code, you pay for in tokens Agentic AI is replaying the cloud's growing pains: mistaking “the machine can do it” for “the machine should do it.” A measurement, and what it taught me.
- What cassation taught me about evaluating AI Law separates what is right from what is justified. Artificial intelligence is rediscovering that line, in disarray.